Using demographics to understand better the students? behavior, improving the performance prediction in online courses

Resumo

Predicting the students' final performance early in term, in online courses, can be an interesting resource to helping them changing their behaviors, aiming to improve their results. A feasible way to do this is applying data mining techniques in their activities' logs. Another important information that should help this analysis, is the students' demographics data, that give us important insights about the expected behavior of each student. However, previous works that tried to use the demographics to improve the performance prediction, did not get so good results. Our study proposes a different approach to use the demographics data, using only this information to generate clusters of students before applying the data mining techniques in the activities' logs, trying to group the students with more similar needs, increasing the chance to get better results. To evaluate the benefits of our method, we compared the accuracy gain in the prediction of the students’ performance, using or not our approach. As results, using the demographics to generate cluster as the first step, and testing with four different methods commonly applied to this purpose, we had an average gain of 11.1% in the accuracy of the predictions.